Can ChatGPT Transcribe Audio? How Files, Record Mode and the API Work

ChatGPT vs VoiceScriber vs OpenAI API: quick comparison

1. Can ChatGPT transcribe audio?

Yes. In 2026, ChatGPT and OpenAI transcription fall into four different buckets:

1. ChatGPT Dictation: Records a spoken prompt, sends the audio to OpenAI for transcription, and returns editable text before you send the message.
2. ChatGPT Record mode: Records, transcribes, and summarizes meetings, brainstorms, interviews, and voice notes. It is currently available in the ChatGPT macOS desktop app for Plus, Pro, Business, Enterprise, and Edu workspaces.
3. OpenAI Speech-to-Text API: A developer API for completed audio files. OpenAI now recommends gpt-transcribe for general file transcription, with specialized models for speaker labels, timestamps, subtitles, and translation.
4. Realtime transcription: An API path for live audio from a microphone, call, or media stream. OpenAI's current Realtime transcription guide recommends gpt-live-transcribe for ongoing streams.

ChatGPT Voice is different from Dictation: it is a back-and-forth conversation, and its transcripts are not verbatim. Playing an existing recording into Voice is not a documented substitute for importing the original audio file.

Can I upload an audio file directly to ChatGPT?

OpenAI's current ChatGPT file-upload pages do not list audio formats such as MP3 or WAV as supported conversation attachments. That does not rule out account-specific or experimental behavior, but it means ordinary audio-file upload is not a documented general ChatGPT workflow. For a completed recording, use the Speech-to-Text API. For speech captured inside ChatGPT, use Dictation or Record mode.

None of these are offline iPhone transcription. If you need to transcribe a private recording on your iPhone without uploading it, ChatGPT is the wrong tool. Use an offline iPhone app like VoiceScriber instead.

2. Record Mode vs OpenAI API vs Realtime: what's the difference?

ChatGPT subscriptions and API usage are billed separately. Speaker-labeled output uses gpt-4o-transcribe-diarize; timestamp and subtitle formats depend on the model. Check OpenAI's model-specific output options. For an iPhone recording you do not want to upload, choose VoiceScriber instead.

3. What data does OpenAI delete, retain, or use for training?

This is the most important privacy question, and the answer depends on which OpenAI product you use.

ChatGPT Dictation

• Audio messages are sent to OpenAI's models for transcription and retained while the chat remains in your history.
• When you delete the chat, OpenAI says the associated audio is deleted within 30 days, subject to security, legal, and previously shared or de-identified data exceptions.
• Sharing Dictation audio for training is opt-in. The transcript text has separate model-improvement controls; disabling audio sharing alone does not opt text out. See the Dictation data controls. Business-user inputs and outputs are excluded from training by default.

Record Mode (ChatGPT macOS app)

• Audio recordings are used only for transcription and deleted afterward.
• Transcripts and canvases follow your workspace retention settings. If you delete the conversation, the associated transcript/canvas is removed from OpenAI systems within 30 days, unless OpenAI is legally required to retain it.
• Reference record history: When enabled, ChatGPT can reference past recording canvas and transcripts in future conversations. This can be useful for meeting memory, but it is a privacy consideration for sensitive topics.
• Consumer ChatGPT: If model improvement is enabled, transcripts and canvases may be used to improve OpenAI models unless you opt out.
• Business, Enterprise, and Edu: Record Mode transcripts and canvases are excluded from model training by default.

Speech-to-Text API

• The /v1/audio/transcriptions and /v1/audio/translations endpoints currently show no training, no abuse-monitoring retention, and no application-state retention.
• OpenAI says API inputs and outputs are not used for training by default unless an organization explicitly opts in.
• These are endpoint-specific content-retention entries, not a zero-storage promise for an entire workflow. Standard /v1/realtime has 30-day abuse-monitoring retention. Your app logs, file storage, and subsequent summary requests need separate review.

Legal and security exceptions

OpenAI's current retention guidance says deletion timelines can be extended for security or legal reasons. Where a file is saved in Library, deleting its chat does not delete that Library copy; remove it separately. This is a storage rule, not evidence that general MP3/WAV attachment support exists.

4. What changed in 2026?

As of October 4, the existing Record and API limits still apply. A newer Meetings plugin adds another cloud workflow, not on-device transcription.

• Record Mode remains a 4-hour / 240-minute feature. Older 120-minute references are outdated.
• Record Mode is still macOS-only. It is not an offline iPhone transcription app.
• Dictation is the direct consumer route for a short spoken prompt. It returns editable text before the message is sent, but the audio is processed in OpenAI's cloud.
• Audio files are best handled through the API, not normal ChatGPT document uploads. OpenAI's ChatGPT file-upload docs focus on documents, spreadsheets, presentations, images, PDFs, and text files, while OpenAI's audio-file transcription docs point to the Speech-to-Text API with a 25 MB audio upload limit.
• The current API recommendations still apply. OpenAI recommends gpt-transcribe for completed files and gpt-live-transcribe for ongoing streams. gpt-4o-transcribe-diarize remains the specialized option for speaker-labeled output.
• Privacy settings are more important. "Reference record history" can let ChatGPT use past recording notes and transcripts in future conversations when enabled.

The separate Meetings plugin is in macOS beta for Pro and Business; Enterprise access is limited alpha. It streams audio to OpenAI. If offline, audio can wait locally until upload and processing finish, so local capture does not mean offline transcription. Its access rules are separate from older Record mode.

For offline iPhone notes, VoiceScriber 1.7.7, released September 14, adds Lock Screen and Dynamic Island recording controls, plus printing notes to paper or PDF. Transcription still runs on-device. See what users say.

5. ChatGPT transcription limits in 2026

For a short spoken prompt, ChatGPT Dictation records your voice and returns editable text. It is not the same as attaching a completed recording, and its audio follows the retention rules for the chat.

This distinction matters because searchers often mix up ChatGPT file upload limits with OpenAI audio transcription upload limits. The ChatGPT document upload limit is not the same thing as the Audio API's 25 MB transcription limit.

For files larger than 25 MB, OpenAI recommends splitting audio into smaller chunks or using compressed formats. To see whether a file is over 25 MB and how long it runs, open it in our free audio length checker, which reads it in your browser without uploading it. If you want a tool that does not depend on API upload limits, Wi‑Fi, or server-side processing, use our offline transcription verification guide for local recording, imports and exports.

6. ChatGPT transcription: key pros

• End-to-end convenience: Record, transcribe, and auto-summarize in one place; outputs can be turned into tasks, emails, or plans.
• Speaker diarization: Record Mode supports multiple speakers; the API offers gpt-4o-transcribe-diarize for speaker identification.
• Enterprise-friendly controls: Business/Enterprise/Edu content is excluded from training by default; admins can control or disable Record Mode.
• Endpoint-specific retention controls: The transcription and translation endpoint table lists no training, abuse-monitoring content retention, or application-state retention. That does not cover every API endpoint or third-party app.
• Multilingual: The speech-to-text models support a wide range of languages, though accuracy varies.

7. Where ChatGPT transcription breaks down

• It is not offline: Record Mode and the API require internet. If you need to transcribe on a flight, in a hospital dead zone, in court, or during field interviews without signal, ChatGPT is the wrong fit.
• Record Mode is not on iPhone: It remains macOS-only. ChatGPT Dictation on iPhone is a different, cloud-processed prompt feature; VoiceScriber is the local recording-and-import alternative.
• Record history can create context bleed: When "reference record history" is enabled, past transcripts can influence future responses. That is useful for continuity but risky for sensitive meetings.
• Consumer training settings matter: Consumer ChatGPT content may be used for model improvement unless you opt out. Many casual users never review this setting.
• Legal holds can override deletion promises: Even when a product has normal deletion windows, active litigation or legal obligations can change what must be retained.
• API file limits remain: The API caps file uploads at 25 MB, and long recordings may need chunking.

8. Who should choose an offline iPhone transcription app instead?

Choose offline transcription when you need control over where a recording is processed, not just a transcript or AI summary.

• Lawyers and legal teams: Attorney-client conversations should stay under your control. A local-only iPhone transcription app reduces third-party exposure. See our guide on secure offline transcription for lawyers.
• Clinicians and therapists: Patient notes and therapy recordings can include sensitive health information. Keeping audio on-device simplifies the data-flow conversation. Read more about on-device transcription for healthcare and therapy notes.
• Journalists and researchers: Source interviews are often sensitive. Offline transcription avoids creating a cloud copy of raw audio. See how offline transcription protects interview sources.
• Finance, HR, and compliance teams: Internal investigations, earnings prep, employee notes, and regulated conversations should minimize unnecessary processors.
• Students, creators, and travelers: Lectures, ideas, and interviews should still be transcribable on a train, plane, or in low-signal environments.

If you would rather not install anything, there are free audio-to-text converters that never upload your file, and our roundup of the best offline transcription apps covers the paid options. To check your own device first, run the offline transcription checker.

Local transcription avoids sending audio to a transcription server, but it does not secure every backup, export, or later AI prompt. For clinical, legal, or regulated use, obtain the required recording consent and organizational approval; on-device processing alone is not a compliance guarantee.

9. The private offline alternative: VoiceScriber for iPhone

If your requirement is "transcribe audio on my iPhone without uploading it", choose a tool built around that requirement from the start.

VoiceScriber is built for high-privacy iOS transcription:

• 100% offline, on-device transcription: Works in airplane mode and does not send recordings or transcripts to a server.
• Private iPhone voice notes: Record, transcribe, edit, search, and export from your device.
• Import existing audio files: Pick a locally available M4A, MP3, or WAV file and transcribe it on-device. OpenAI's 25 MB API cap does not apply; each imported file can run up to 90 minutes.
• 100+ languages supported offline.
• No account required: No sign-up barrier, cloud sync requirement, or server-side transcript storage.
• Up to 90 minutes per recording: This is a recording limit, not a promised free allowance. Current product facts list iOS 17.6 or later, free-to-try access, $5.99 weekly, and $49.99 lifetime options; country pricing may vary.
• Built for real-world privacy workflows: Useful for lawyers, clinicians, journalists, students, creators, and teams that need transcription without Wi‑Fi.

10. At-a-glance: ChatGPT transcription vs. VoiceScriber

11. If you must use ChatGPT for transcription, harden your setup

1. Turn off training for consumer ChatGPT: Use OpenAI's data controls so future conversations are not used to improve models.
2. Check your organization's approved service: Business offerings and API data are excluded from training by default, but that alone does not approve a sensitive or regulated workflow.
3. Review "Reference record history": Disable it if you do not want past recording transcripts and canvases referenced in future chats.
4. Check the exact endpoint and storage: Transcription/translation policies do not cover your app's logs, stored files, Realtime sessions, or a later summary request.
5. Do not mix sensitive recordings with general prompts: Keep privileged, clinical, financial, or HR recordings separate from casual ChatGPT use.
6. Choose offline for zero-upload needs: If policy says audio must not leave the device, use VoiceScriber on iPhone, then review setup, imports, exports, and backups against that policy.

Related articles

• VoiceScriber vs. Cloud Transcription
• 5 Reasons to Choose Offline AI Transcription (Privacy, Speed & Cost)
• Transcribe in 100+ Languages Offline on iPhone
• How to Transcribe Meetings & Lectures on iPhone Completely Offline
• Secure Offline Transcription for Lawyers
• 5 Best Otter.ai Alternatives (Free, Offline & Private)
• Which Transcription Apps Work Offline? Verification Guide
• SOAP and Therapy Notes on iPhone: On-Device Dictation
• Best Offline Transcription Apps by Platform and Workflow
• Wispr Flow Pricing 2026: Free Plan, Pro Cost, Student Discount & Offline iPhone Alternative
• Best Voice-to-Text Apps for iPhone in 2026: Privacy, Offline Mode & Price
• How to Transcribe Audio Offline for Free With Whisper on Windows, Mac & Linux

FAQs

Sources

Research checked October 4, 2026.

• OpenAI: ChatGPT Record, Meetings plugin, Dictation, and ChatGPT Voice.
• OpenAI: File transcription, Realtime transcription, and API data controls.
• OpenAI: Supported file types, File Uploads FAQ, and Chat and file retention.
• VoiceScriber: Product facts and App Store listing and release history.